Alex Good on Bitcoin's weakening thesis, proof-of-work AI coins, and how AI is changing market cycles

Alex Good on crypto's direction, proof-of-work AI coins, and AI-accelerated markets

Alex Good built Post Fiat after stints at Citigroup, Palantir, and Baly Asmi, and an exit from a company he founded. His framing is that AI was always going to destroy most knowledge work — he says he arrived at that conclusion in 2023, well before it became consensus — and his current project applies what he calls Palantir-style ontology methods to aggregate decentralized trading intelligence and generate market alpha.

Bitcoin's weakening position

Good argues the Bitcoin thesis is getting weaker by the day. The core problem is two-sided. First, privacy coins are gaining real traction: EU attempts to kill Monero failed, and US Tornado Cash rulings have given legal clarity to coins like Railgun, which Vitalik Buterin uses. Bitcoin, by contrast, is not private technology. Second, Michael Saylor's leverage concentration — preferred stock and complex convertibles stacked on top of Bitcoin — has tightened the liquidity profile in ways Good sees as structurally hard to reverse. He points to the MSTR equity situation and the STRC drop as live examples.

The underlying thesis keeping him in crypto at all is what he calls the "doom thesis" — that AI may accelerate the surveillance state rather than productivity, which drives capital flight into privacy-preserving assets. Peter Thiel relocating to Argentina is, in his read, a signal that this dynamic is already in motion.

“The Bitcoin thesis is getting weaker by the day. You have privacy coins accelerating in terms of their adoption. Bitcoin is fundamentally not private technology, and it has unfortunate leverage concentration from Michael Saylor. Pearl is a proof-of-work AI coin trading at a $2,000,000,000 FDV on private — it's a regime shift. The way that people interact with markets has changed: they're now pasting earnings transcripts into AI systems, and those AI systems have biases quite separate from the biases humans have.”

Proof-of-work AI coins

The most forward-looking claim Good makes concerns a project called Pearl, which he says is trading at a $2 billion FDV on a private token sale and is backed by Thrive Capital (described as rumored). Pearl is an Israeli team using MapMul mining in partnership with TogetherAI, allowing AI computation to secure a cryptocurrency — a structure called useful proof-of-work.

The architecture is meaningfully different from earlier crypto-compute networks like Render or BitTensor, which Good says are stake coins not secured by actual GPU operation. Pearl's model means miners subsidize inference costs: Good says users can buy Gemma on Together.ai at a 25% discount because miners are subsidizing it. He immediately flags the limit — OpenRouter already offers the same model at roughly 50% below Pearl's subsidized price, so the value proposition is market-cap dependent. The coin needs to appreciate materially for the subsidy to mean anything.

The broader claim is a TSMC data point: in 2017-18, Bitcoin ASICs represented roughly 10% of TSMC's output. Today it's well under 1%. If you want to secure money with proof-of-work, the question is why you wouldn't use GPUs instead of purpose-built ASICs. Good acknowledges the counterargument — burning productive GPUs to secure a coin is, in some ways, worse than burning energy for Bitcoin mining — but treats it as a secondary objection.

CoreWeave is the reference case for the compute-to-AI pivot: Good describes it as "the OG pivot story," having moved out of crypto mining into AI GPU clouds before the JCAL "pivot to AI" post in 2023.

AI changing who moves markets

Good's read on market cycles is that AI has changed the composition of engaged investors more than it has changed the cycle itself. Retail investors now paste earnings transcripts into Claude or GPT and act on the output. The models, he argues, have different biases than humans — Claude finds ASML's earnings call genuinely interesting in a way most human listeners wouldn't, which means esoteric industrial and semiconductor stories are getting retail attention they never would have before. He names HPE, Micron, and SanDisk as examples: stocks historically held by 70-year-olds as bellwethers, now up 300% on AI infrastructure narratives, while Coinbase shows substantial sequential revenue deceleration as retail rotates toward B2B infrastructure plays.

He applies the same logic to crypto: coins with complex technical stories — Zcash's encryption stack, for instance — now get engagement because the friction of understanding them has collapsed. Lore matters less; technical depth matters more, because Claude can explain it.

AI in entertainment: deferred, not canceled

Good is skeptical of near-term AI disruption to gaming. Real-time generative NPCs would cost roughly $6 per game session at current compute prices, which he says makes unit economics impossible. Nintendo is down roughly 40% over the past year; Take-Two is getting "smoked" despite GTA. The intuitive bull case — AI makes game development dramatically cheaper or enables fully generative content — is structurally blocked by cost for now.

His longer-run view is the opposite of bearish. As AI inference costs fall, entertainment becomes a major beneficiary precisely because frontier labs have deprioritized it. Anthropic has explicitly said it won't generate video or images; OpenAI shut down Sora. That structural crowding-out of entertainment AI creates space for companies like Suno. When costs come down, Good expects a significant acceleration in AI-generated entertainment.

Getting out of AI psychosis

Good describes a convergence pattern for people who go deep on AI's implications: initial panic that the work is meaningless, followed by direct engagement with the tools, followed by the observation that AI is strong in areas you don't deeply understand and weak in areas you do. His exit from the psychosis was building Post Fiat — using AI to aggregate contributions from hundreds of people in the Microbare Project "data lake," summing their trading intelligence into actionable ideas. The framing is that AI doesn't eliminate edge; it enables new data primitives that weren't economically feasible to build before.